Vehicle model based outlier detection for automotive visual odometry

In this paper we present a novel outlier detection scheme for image feature based ego-motion estimation in automotive applications. It is based on a restrictive motion model, describing the relationship between road vehicle motion and camera motion. The model also enables the integration of ESP sensor data, such as measured longitudinal velocity and yaw-rate. In this way a high precision camera motion prediction is realized, which is used to identify erroneous feature correspondences. High costs of standard methods like the iterative random sample consensus (RANSAC) [1] are thereby avoided.